Night security drones pair a thermal sensor for guaranteed detection with a starlight or NIR-assisted camera for identification — and AI tracking to hold a moving person or vehicle automatically along a perimeter. The two channels are not redundant: thermal tells you someone is there, and only the low-light channel tells you who.
- 0.001 luxStarlight sensor class
- PassiveThermal emits nothing
- InvisibleNIR illumination band
- 1 operatorWith AI tracking
Key takeaways
- Thermal detects, low-light identifies. A thermal image shows a warm human shape with no clothing, face or object detail — enough to dispatch, not enough to describe.
- Thermal is fully passive. Unlike a floodlight, patrol does not announce itself or reveal its timing.
- NIR illumination extends the low-light channel into true zero light while staying invisible to the naked eye.
- AI tracking is what makes one operator able to run a patrol. Without it, the gimbal is a full-time job.
On this page
Detection vs identification at night
Thermal detects any warm body regardless of light but renders no facial or clothing detail. Starlight sensors deliver full-colour imagery at very low lux; NIR laser illumination extends usable identification range in true zero light. Serious patrol payloads carry both.
The operational consequence is that a thermal-only patrol generates contacts it cannot resolve. “A person is at the north fence” dispatches a guard; “a male in dark clothing carrying a bag is at the north fence” briefs one. On sites where response is by a small team, that difference determines whether the patrol reduces workload or increases it.
The three night technologies do different jobs and are complementary rather than competing.
| Channel | Works in | Gives you | Range class | Passive? |
|---|---|---|---|---|
| Thermal (LWIR) | Total darkness, light smoke | Presence, shape, movement | Longest | Yes |
| Starlight CMOS | Starlight, ~0.001 lux | Full colour, clothing, faces at range | Medium | Yes |
| Starlight + NIR laser | True zero light | Monochrome detail at distance | Medium–long | No (invisible) |
| Visible + spotlight | Any | Full detail | Short | No (obvious) |
| Thermal + starlight fusion | Darkness | Detection with context | Long | Yes |
| AI detection layer | Any of the above | Automatic flagging of person/vehicle | — | — |
Do not reach for a spotlight out of habit. On a thermal patrol it adds nothing to the thermal channel and saturates the low-light channel it was meant to help. Establish the sequence explicitly: detect on thermal, identify on low-light, illuminate only when a colour identification is genuinely required.
AI tracking on patrol
Human and vehicle detection flags targets automatically during route flights; click-to-track keeps the gimbal locked while the aircraft repositions. This converts a camera into a patrol tool one operator can actually run.
The workload argument is the whole point. A patrol leg flown manually requires the operator to fly the aircraft, scan the sensor and interpret what they see, simultaneously, for the duration. Automatic flagging reduces that to monitoring an alert stream; click-to-track removes the gimbal from the workload entirely once a contact is held.
Be realistic about failure modes before relying on it operationally. Occlusion, rapid scale change and thermal crossover at dawn and dusk all defeat trackers, and the recovery behaviour differs by implementation. The specifications that predict operational quality are tracking deviation refresh rate and output delay rather than raw detector frame rate — see onboard AI inference architecture.
Matching zoom reach to perimeter size
The payload should be chosen from the geometry of the site, not from a specification sheet.
A compact industrial site with a 400 m perimeter can be watched from a single orbit, and identification range of a few hundred metres is sufficient. A distribution yard or a long fence line needs far more reach, because the aircraft cannot be everywhere and identification has to happen from wherever it is. Work backwards: decide the orbit positions the site allows, measure the longest identification distance that geometry forces, then select the optical channel that delivers it.
Over-specifying costs real money and real endurance. A long-reach pod on a small site adds mass, reduces flight time and buys reach nobody uses. Under-specifying produces a patrol that detects everything and identifies nothing.
What a night patrol programme gets wrong
- Thermal-only payloads on identification-driven sites. Generates contacts the team cannot action.
- Using visible illumination during thermal search. Saturates the low-light channel and announces the patrol.
- Trusting AI tracking through occlusion. Test the recovery behaviour before an operational dependency forms.
- Ignoring dawn and dusk crossover. Contrast collapses for roughly twenty minutes; schedule around it.
- No retention or access policy. Perimeter footage contains people, including staff and neighbours. Decide policy before the first flight — see the privacy and regulation primer.
- No coordinate on the contact. A track that reaches the response team as a description rather than a position wastes the detection.
Recommended UAVThermal payloads
MV-1P — 100 g black-light full-colour night vision for micro platforms, where a small quiet aircraft matters more than reach.
OP-80N — ultra-starlight zoom with NIR laser illumination for zero-light perimeters that need identification at distance.
MV-4X — starlight, 640×512 thermal and NIR in one 366 g quad-sensor micro payload. Detection and identification without a payload swap.
LX-8RB — long-reach zoom with laser night vision and ranging for wide-area sites and long fence lines.
Related reading
Technology: starlight and low-light imaging and NIR laser illumination.
Field practice: starlight vs NIR vs thermal compared, how AI tracking actually works and night flight lighting and compliance.
Adjacent missions: border and perimeter surveillance and airport perimeter and wildlife.
FAQ
Why do I need both thermal and a low-light camera?
They answer different questions. Thermal detects any warm body regardless of lighting but shows no clothing, face or object detail. A starlight or NIR-assisted channel supplies that detail. A thermal-only patrol generates contacts the response team cannot describe or brief.
Does NIR illumination give away the patrol?
Not to the naked eye — near-infrared illumination is invisible to human vision. It is detectable by other NIR-capable sensors, so it is covert rather than undetectable. Thermal, by contrast, is fully passive and emits nothing at all.
Can one operator run a patrol?
With AI detection and click-to-track, yes. Automatic flagging reduces scanning to monitoring an alert stream and tracking removes the gimbal from the workload. Without those, flying, scanning and interpreting simultaneously is a two-person job on any sustained leg.
How much zoom does my perimeter need?
Work backwards from geometry. Decide the orbit positions the site allows, measure the longest identification distance that geometry forces, then choose the optical channel that delivers it. Over-specifying costs mass and endurance for reach nobody uses.